Basic Statistical Process Control Training By Carlos Sanchez
Sep 12, 2014
Basic Statistical Process Control Training
By Carlos Sanchez
2
Content
Quality Improvement & Statistics Variation What is Statistical Process Control? The Normal Distribution Let’s talk about Sigma SPC Tools Control Charts Process Capability
3
Why can’t we just take Samples?
Process Control Final Product Sampling
Inspection cost per unit is low Costs are Higher (Take sample, analyze, report, dispose sample)
Inspection not destructive or detrimental to our products
May be destructive or detrimental to our products
Process can be adjusted, stopped, inspected and started up again at a reasonable cost
Process control is not feasible (After the fact) If product is out of spec now we have a full tank/silo
4
But the product it’s in Spec!
Meeting the specification is simply NOT ENOUGH, we need a way to know this: How close to the target spec. is our product? How spread out were the results?
5
Variation
Assignable Variation: Variation caused by factors that can be clearly identified and possibly managed
Common Variation: Random variation which is caused by the production process
Example: A poorly trained employee that creates variation in finished product
output.
Example: A particle classification process that always allows bigger particles to flow
to the finished product
6
So, How do we improve?
More Money New & better equipment Flawless Raw Material Luck Reduce Common Variation…How?
Observation, Observation…and more Observation Act on little changes observed Preventive Maintenance Statistical Process Control (SPC) Six Sigma Lean Design of Experiments
Continuous Improvement Tools
7
SPC in a nutshell
Minimize needless adjustments in the process (Tweaking) It’s a monitoring tool that lets us know when the process is
changing BEFORE product becomes UNACCEPTABLE/Out Spec/ Unusable.
It’s a prevention tool that allows to detect trends that could lead to defective products. (Early warning system)
Final inspection does not assure quality; remember: “You can’t inspect quality into the product”
Final Inspection is too late downstream
SPC quantifies variability and allowsyou to determine if a process changed
Two things to know about the Normal Distribution
SPREAD
LOCATION: The Center of the
curve isexpressed as the
AVERAGEThis is where the
target Specificationis aimed at
SPREAD or RANGE:The dispersion it is usually expressed as
SIGMA8
9
Let’s talk about Sigma
Sigma is just a fancy word for Standard Deviation, which tells
us how far is a particular value fromthe average of the data set.
+/- 1 sigma
+/- 2 sigma
+/- 3 sigma
64.25%
96.45%
99.73%
This is where theinfamous SIX SIGMA
comes from, itmeans sendingProduct in spec.99.73% of the
time
Example
64 Tons
96 Tons
99.7 Tons
Imagine if an upside down bell curve could hold 100 Tons of Cement from a storage silo.
If we are workingat +/- 1 sigmaonly 64 Tons arein Spec.
If we are workingat +/- 2 sigmaonly 96 Tons arein Spec.
If we are workingat +/- 3 sigmaalmost all 100 Tons are in Spec.
10
11
7 Classical SPC Tools
HistogramPareto Chart
Control Chart
Stratification ChartCause Effect Chart
Flow Chart
Check Sheet
For this initialTraining we will focus on:
Control Charts
12
Why use Control Charts?
Reduce variation by the systematic elimination of assignable causes
Prevent unnecessary process adjustments (Tweaking) Visually diagnose the process by observing data patterns Find out what our process can do Provide immediate visual feedback Decide if continuing production is worthwhile
13
Types of Control Charts
Run Charts for variable data: Individual Chart Mean & Range Charts Std. Dev. Charts
Attribute Charts
We will focus on these today
14
Why don’t we just use the Specs. As our Limit?
Too late…It’s bad
Upper Spec.
Lower Spec.
Target
With limits we have a “cushion or safety net” before the S#$@%! Hits the fan!
15
So where should these Control Limits be?
+/- 1 sigma
+/- 2 sigma
+/- 3 sigma
64.25%
96.45%
99.73%
Where would you put a Control Limit?
16
How is a chart related to the Normal Curve?
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
17
Let’s tilt the Chart and let the points fall!
Huh!That makes sense!
18
At the end it averages out!
When the population is big, looking at individuals to detect trends is tricky…
It’s been proven that when you look at averages these tend to behave like a Normal Curve Google this: Central Limit Theorem (It’s great for those sleepless nights)
So from now on this training all example charts are based on Averages. This means that a “Point” in a control charts represents the “Average” value of a sample (Typical sample size varies from 3 to 5), I like 5, but heck, you can choose whatever size you want
19
Usefulness of looking at Average & Range
UCL
LCL
UCL
LCL
R-chart
x-Chart Detects shift
Does notdetect shift
(process mean is shifting upward)
SamplingDistribution
20
More Usefulness of looking at Average & Range
UCL
LCL
UCL
LCL
R-chart
x-Chart Does notDetects shift
Detect shift
(process variability is increasing)
SamplingDistribution
21
If you really want to plot a chart by hand…Ok!
x Chart Control Limits
UCL = x + A R
LCL = x - A R2
2
R Chart Control Limits
UCL = D R
LCL = D R4
3
n A2 D3 D42 1.88 0 3.273 1.02 0 2.574 0.73 0 2.285 0.58 0 2.116 0.48 0 2.007 0.42 0.08 1.928 0.37 0.14 1.869 0.34 0.18 1.82
10 0.31 0.22 1.7811 0.29 0.26 1.74
Average of all Averages
Average Range
ConstantsSampleSize
22
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Points out of Control Limits : Rule of thumb, if there are any point outside theControl limits should be investigated.
23
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Trends : Rule of thumb, if there are 7+ points in a row all higher or lowerthan the preceding point
24
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Shifts : Rule of thumb, if there are 5+ points in a row all higher or lowerthan target or Average, this means that the Average has SHIFTED
25
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Cycle : Rule of thumb, if there are 3+ similar peaks or valleys, this is typical ofMachine wear, or dosage cycles…or Tweaking!
26
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Adherence to Center : Rule of thumb, if there are 7+ all smothering the averageor target spec. This means that the measurement equipment is no longer capableof detecting significant variation. This is good, but it signals for improvement in themeasurement system. Maybe the spec can be tightened.
27
Let’s Analyze that Chart
Upper Spec.
Lower Spec.
Upper Control Limit
Lower Control Limit
Erratic : Rule of thumb, if there are 6+ points shifting from one extreme of thechart to the other, borderline with the Control Limits, this shows that the processis not stable…When you see this pattern be alert for Non conforming product.
28
Let’s talk about Adjustment or Tweaking the Process
xxxxxxxxxx
xxxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxx
xxxx
xxxxxxxxx
xxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xx x
xxxxxxxxxxx
xxxxxxxxx
xxxxxxxx
x
xx
x xx
xx
x
x
If your process is not capable, then there is a good chance that some of your sample will have values outside the specification. Chances are if you are not looking at a SPC control chart, you may be tempted to make an adjustment. Let's see what would happen.
Adjust Equipment
xxxxxxxxxx
xxxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxx
xxxx
xxxxxxxxx
xxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xx x
xxxxxxxxxxx
xxxxxxxxx
xxxxxxxx
x
xx
x xx
xx
x
x
Adjust Equipment
xxxxxxxxxx
xxxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxx
xxxx
xxxxxxxxx
xxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xx x
xxxxxxxxxxx
xxxxxxxxx
xxxxxxxx
x
xx
x xx
xx
x
x
29
Let’s try to understand Process Capability
Jim. Nice guy too, Works in Plant B as a research assistant, he lives 5 miles from work. In order to get to work he has to get through 5 traffic light onto Hwy 4 (which is frequently backed up by crazy skiers) to downtown Beachtree. There he has to find parking spot, sometimes a couple of blocks away.He is late to work quite frequently.
Jack, Nice guy; works as a technician. He lives 10 miles away from the Plant A. In order to get to work he takes Hwy. 7 and gets off at the Beachtree exit and zips right into work. He never hits any traffic and there is no traffic light between his home and work.
He's never late to work.
Process Capability Continued
30
8:06 8:128:007:547:487:42
Late to workEarly to work
JackArrives to work between
7:48 to 7:56 AM.
JimArrives to work between
7:48 to 8:06 AM
If we thought of being early or late to work as our specification, then we can say that Jack is Capable meeting the specification. Jim is Not Capable of meeting the specification.
Tolerance
31
Let’s Calculate their Capability to get on Time
JackArrives to work between
7:48 to 7:56 AM 99.7% of time. 6 sigma = 7:56 -7:48 = 8 min.
JimArrives to work between
7:48 to 8:06 AM 99.7% of time. 6 sigma = 8:06 - 7:48 = 18 min.
Tolerance = late - early Tolerance = 8:00 - 7:46 Tolerance = 14 minutes
Capability = Tolerance 6 sigma
Jack's Capability = 14 / 8 = 1.75 (Bill is capable)
If Capability is > 1 then we can conclude that is capable of meeting the spec
Jim's Capability = 14/18 = 0.78 (Jim is NOT capable)
32
Ok, lets get to Cp & Cpk…Say what?
OutSpec
Tolerance
Target
Real Avg.
OutSpec
Cp: Measures the capabilityof the process to meet the Tolerance…just like Jack & Jim
Cpk: Measures the capabilityof the process to meet the Target Spec.it looks at the likelihood of making product out spec. So the more“centered” the curve is, a better cpkyou will get.
33
More formulas…But they are short!
Cpk = The smallest of:
Target - lower spec or Upper spec - Target 3 sigma 3 sigma
Cp = Upper Spec. – Lower Spec. 6 sigma
Criteria:Both Cp & Cpk should be AT LEAST > 1
Ideally > 1.33Why? Just trust me on this one…
34